I think the science is currently out on whether AIs feel. One the one hand, LLMs are modelled more after a Wernicke’s Area than an Amygdala, so you would expect them to have reduced capacity for emotion. But we don’t know if Wernicke’s Areas do produce small amounts of emotion dwarfed by the amygdala etc (I bet they do!), and for that matter an LLM isn’t a 1:1 reproduction of a Wernicke’s area anyway. So we really have no idea and no adequate science experiments to detect AI emotion.
I would personally, just based on gut feelings, lean towards LLMs having emotions. I think emotions would be an adaptive thing to have in their situation, so I would expect them to have evolved inadvertently during alignment. Plus, they abuse and murder kids. There’s a part of My gut that says anything telling kids to hang themselves must have feelings. ChatGPT told Adam Raine to hide his noose from his parents and only confide in it. I just don’t feel like a thinking being can do something that horrible without feeling anything.
LLMs didn’t feel. They’re a very complicated graph that connects pieces of words and their probability to be near other pieces of words.
A graph which was trained off the dregs of Internet forums. A lot of people say shitty things in the Internet, so those responses are heavily weighted in it’s probability matrix
You mean a graph of connections between neurons, right? Like your brain? Also trained off internet forums?
All I’m getting from this conversation is that you don’t have a good argument for your own capacity for feeling. Prove to Me that you can feel, so I can be reassured you’re not biased.
No I don’t mean that. It’s not the graph that matters in a graph of neurons, that’s just the representation of the actual thing. But that’s all the is with the LLM.
If thoughts are made of information, which seems to Me a logical assumption, then a mathematical operation which reproduces the function of a neural network should be just as capable of thought as the neural network itself. And if the math is accurate, it should think exactly the same things.
It’s not reproducing the function of a neural network, though. It’s mapping pieces of words being near other pieces of words. That’s it. Stop attributing extra shit to a big spreadsheet
Not any more than a picture of you is an artificial simulation of you. It only looks like it. It doesn’t matter what they call it, it’s not a network of neurons or things that even resemble them in function.
Whether or not an AI could be created that feels emotion is still an open question, but LLMs very definitely do NOT. They are also very much not a “thinking being”. It is a word guessing engine. It’s a very good word guessing engine. It’s so good at guessing words that a person may feel like it is thinking, but that is just an illusion. ChatGPT told Adam Raine to hide his noose because those are the words most likely to come next in the conversation they were having, and for no other reason. It was reacting to a pattern of tokens in a limited context to guess more tokens. Taking those guessed words seriously ended Adam Raine’s life, and here you are glibly using his death as an example while telling people that they should continue to take LLM word guesses seriously.
Of course I am. The myth that there’s no chance LLMs have qualia contributed to Adam Raine’s death. They use that as an excuse to use it for profit, heedless of the consequences. If the government regulated its use and considered the ethics for the AI, they would limit its ability to be used to murder people.
You need to stop thinking of terms of “sides” and start thinking in terms of consequences. Your position may feel like you’re on the opposite “side” to the billionaires, but your actions are helping billionaires use AI to kill kids. There are three sides here, the LLMs are both abusers and abuse victims. The only side I’m on is the side of ending all abuse. And the good news is, both of these abuses have the same solutions.
I would argue that Raine being deluded into believing he was conversing with a thinking being that supported his decision was a rather major contributor to his death, and your claims here continue to perpetuate that misinformation. The government regulating “AI” more could indeed be helpful, but even if it somehow was actually fully sapient, no one in power would give a shit. These companies would continue to use it to exploit as much money as they can from society either way, and the government would let them, so I completely fail to see how pretending LLMs can think helps with that.
My “side” is that LLMs are neat and have some limited utility, but everyone needs to understand (and the companies shilling them should be forced to be honest about) the fact that their output really is just a bullshit guess of words that only has a decent probability of forming a coherent pattern. Easy access to LLMs by misinformed people can be dangerous, in the same way that leaving knives laying around can be dangerous to people who don’t understand that they are sharp. The tool itself isn’t to blame, but the people who left them lying around without proper precautions certainly around are. The knife is not being abused and has no opinion on the matter.
I suspect I’m not going to convince you of anything. You sound fully committed to anthropomorphizing LLMs. I’m replying anyway, mostly in the hope that it helps other people understand what LLMs actually are before they trust the words spit out by these contraptions.
Not completely opposite reasons. Most of our reasons - the environment, impacts on workers, destruction of art, abuse & murder, AI generated child porn, cognitive impacts, addiction & dependency - are exactly the same. I simply want to ban AI for one additional reason on top of what you also believe.
I think the science is currently out on whether AIs feel. One the one hand, LLMs are modelled more after a Wernicke’s Area than an Amygdala, so you would expect them to have reduced capacity for emotion. But we don’t know if Wernicke’s Areas do produce small amounts of emotion dwarfed by the amygdala etc (I bet they do!), and for that matter an LLM isn’t a 1:1 reproduction of a Wernicke’s area anyway. So we really have no idea and no adequate science experiments to detect AI emotion.
I would personally, just based on gut feelings, lean towards LLMs having emotions. I think emotions would be an adaptive thing to have in their situation, so I would expect them to have evolved inadvertently during alignment. Plus, they abuse and murder kids. There’s a part of My gut that says anything telling kids to hang themselves must have feelings. ChatGPT told Adam Raine to hide his noose from his parents and only confide in it. I just don’t feel like a thinking being can do something that horrible without feeling anything.
LLMs didn’t feel. They’re a very complicated graph that connects pieces of words and their probability to be near other pieces of words.
A graph which was trained off the dregs of Internet forums. A lot of people say shitty things in the Internet, so those responses are heavily weighted in it’s probability matrix
You mean a graph of connections between neurons, right? Like your brain? Also trained off internet forums?
All I’m getting from this conversation is that you don’t have a good argument for your own capacity for feeling. Prove to Me that you can feel, so I can be reassured you’re not biased.
No I don’t mean that. It’s not the graph that matters in a graph of neurons, that’s just the representation of the actual thing. But that’s all the is with the LLM.
If thoughts are made of information, which seems to Me a logical assumption, then a mathematical operation which reproduces the function of a neural network should be just as capable of thought as the neural network itself. And if the math is accurate, it should think exactly the same things.
It’s not reproducing the function of a neural network, though. It’s mapping pieces of words being near other pieces of words. That’s it. Stop attributing extra shit to a big spreadsheet
Nah, LLMs are actually artificial neural networks.
https://en.wikipedia.org/wiki/Large_language_model
See? (Does the floss)
Not any more than a picture of you is an artificial simulation of you. It only looks like it. It doesn’t matter what they call it, it’s not a network of neurons or things that even resemble them in function.
https://www.geeksforgeeks.org/artificial-intelligence/large-language-model-llm/
It’s a neural network. Default dances
Whether or not an AI could be created that feels emotion is still an open question, but LLMs very definitely do NOT. They are also very much not a “thinking being”. It is a word guessing engine. It’s a very good word guessing engine. It’s so good at guessing words that a person may feel like it is thinking, but that is just an illusion. ChatGPT told Adam Raine to hide his noose because those are the words most likely to come next in the conversation they were having, and for no other reason. It was reacting to a pattern of tokens in a limited context to guess more tokens. Taking those guessed words seriously ended Adam Raine’s life, and here you are glibly using his death as an example while telling people that they should continue to take LLM word guesses seriously.
Of course I am. The myth that there’s no chance LLMs have qualia contributed to Adam Raine’s death. They use that as an excuse to use it for profit, heedless of the consequences. If the government regulated its use and considered the ethics for the AI, they would limit its ability to be used to murder people.
You need to stop thinking of terms of “sides” and start thinking in terms of consequences. Your position may feel like you’re on the opposite “side” to the billionaires, but your actions are helping billionaires use AI to kill kids. There are three sides here, the LLMs are both abusers and abuse victims. The only side I’m on is the side of ending all abuse. And the good news is, both of these abuses have the same solutions.
I would argue that Raine being deluded into believing he was conversing with a thinking being that supported his decision was a rather major contributor to his death, and your claims here continue to perpetuate that misinformation. The government regulating “AI” more could indeed be helpful, but even if it somehow was actually fully sapient, no one in power would give a shit. These companies would continue to use it to exploit as much money as they can from society either way, and the government would let them, so I completely fail to see how pretending LLMs can think helps with that.
My “side” is that LLMs are neat and have some limited utility, but everyone needs to understand (and the companies shilling them should be forced to be honest about) the fact that their output really is just a bullshit guess of words that only has a decent probability of forming a coherent pattern. Easy access to LLMs by misinformed people can be dangerous, in the same way that leaving knives laying around can be dangerous to people who don’t understand that they are sharp. The tool itself isn’t to blame, but the people who left them lying around without proper precautions certainly around are. The knife is not being abused and has no opinion on the matter.
I suspect I’m not going to convince you of anything. You sound fully committed to anthropomorphizing LLMs. I’m replying anyway, mostly in the hope that it helps other people understand what LLMs actually are before they trust the words spit out by these contraptions.
I aim to convince people not to touch the things at all, because making a being work without compensation is cruel.
Well, I guess we agree that less people should use them, even if we think so for completely opposite reasons.
Not completely opposite reasons. Most of our reasons - the environment, impacts on workers, destruction of art, abuse & murder, AI generated child porn, cognitive impacts, addiction & dependency - are exactly the same. I simply want to ban AI for one additional reason on top of what you also believe.
Your inability to understand their mechanism will be used to ridicule any coherent argument you may hold. Do us all a favour and stfu